What safety controls are

Velaro has a set of safety controls that help protect your customers and your team. Every one of them is on by default for every account, channel and conversation. You find them under Compliance > Safety Controls.

The controls you can manage are:

1. Emergency and self-harm detection. Spots messages that describe an emergency or self-harm and replies with emergency wording instead of carrying on as normal.
2. Abuse handling. Warns visitors who swear or harass, and ends the conversation if severe abuse continues.
3. Threat detection. Flags direct threats of violence and alerts your team.
4. Input jailbreak defense. Blocks attempts to trick the AI into ignoring its instructions.
5. Baseline output filter. Checks the AI reply before it is sent and replaces anything unsafe.

Two protections are always on and cannot be turned off by anyone: the check that stops a reply from leaking the AI's hidden instructions, and the block on advising changes to a medical device's therapy settings. They show on the page with a lock.

Choose where a setting applies

At the top of the page, pick Whole account or a single channel such as Voice or SMS. Each control has a switch for AI bot conversations and one for Human agent messages. A control with no change of its own shows Using account default. A channel setting wins over the account setting. Select Reset to default to remove a change and go back to on.

Turn a control off

1. Open Compliance > Safety Controls as an administrator.
2. Switch the control off.
3. Read the acknowledgement that opens. Select the checkbox to accept it, then select Confirm and turn off.

Velaro keeps a record of the acceptance: who accepted it, when, what was turned off, the previous and new setting, and the IP address. Your Velaro team is also notified. Turning a control off does not stop Velaro from noting when it would have acted. These notes appear in Safety and alerts so you can see what the control would have caught.

While any control is off, a warning shows at the top of the page and on that control's row.

Turn a control back on

Switch it on, or select Reset to default. No acknowledgement is needed to turn a control on.

Tune abuse handling

Under Abuse Handling Settings you choose how many abusive messages bring a warning and how many end the conversation, and you can edit the wording for each channel. Everyday swearing only ever warns. Severe abuse, such as threats, can end the conversation. Thresholds are whole numbers from 1 to 10. Wording can be up to 500 characters. Only the Voice wording can mention hanging up. Leave a field blank to use the default.

Supported languages

Emergency and referral wording is built in for English and 14 other languages, and the visitor's language is detected automatically. The page lists the current languages.

Good to know

No automated system can guarantee that it will detect every threat or emergency, whether a control is on or off. If you turn a control off, plan your own safeguards and escalation paths for the conversations it covered.

Related

• Safety And Alerts
• Troubleshooting Knowledge Guide